List of usage examples for org.apache.mahout.vectorizer.common PartialVectorMerger NO_NORMALIZING
float NO_NORMALIZING
To view the source code for org.apache.mahout.vectorizer.common PartialVectorMerger NO_NORMALIZING.
Click Source Link
From source file:com.caseystella.ingest.SparseVectorsFromSequenceFiles.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option libJarsOpt = obuilder.withLongName("libjars") .withArgument(abuilder.withName("libjars").withMinimum(1).withMaximum(1).create()) .withDescription("The default arg for libjars").withShortName("libjars").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(/* ww w . j a v a 2 s .co m*/ "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) of the document frequencies of these vectors." + " Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less then 0 no vectors " + "will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(libJarsOpt).withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt) .withOption(minDFOpt).withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt) .withOption(powerOpt).withOption(minLLROpt).withOption(numReduceTasksOpt) .withOption(maxNGramSizeOpt).withOption(overwriteOutput).withOption(helpOpt) .withOption(sequentialAccessVectorOpt).withOption(namedVectorOpt).withOption(logNormalizeOpt) .create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = DefaultAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it ClassUtils.instantiateAs(analyzerClass, Analyzer.class); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); //TODO: move this into DictionaryVectorizer , and then fold SparseVectorsFrom with EncodedVectorsFrom to have one framework for all of this. DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; if (!processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); long vectorCount = docFrequenciesFeatures.getFirst()[1]; maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:com.digitalpebble.behemoth.mahout.SparseVectorsFromBehemoth.java
License:Apache License
public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option typeNameOpt = obuilder.withLongName("typeToken").withRequired(false) .withArgument(abuilder.withName("typeToken").withMinimum(1).withMaximum(1).create()) .withDescription("The annotation type for Tokens").withShortName("t").create(); Option featureNameOpt = obuilder.withLongName("featureName").withRequired(false) .withArgument(abuilder.withName("featureName").withMinimum(1).withMaximum(1).create()) .withDescription(//from w w w . j a v a 2 s.c om "The name of the feature containing the token values, uses the text if unspecified") .withShortName("f").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription( "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) of the document frequencies of these vectors." + " Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less then 0 no vectors " + "will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option labelMDOpt = obuilder.withLongName("labelMDKey").withRequired(false) .withArgument(abuilder.withName("label_md_key").create()) .withDescription("Document metadata holding the label").withShortName("label").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(typeNameOpt) .withOption(featureNameOpt).withOption(analyzerNameOpt).withOption(chunkSizeOpt) .withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt).withOption(maxDFSigmaOpt) .withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt).withOption(minLLROpt) .withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt).withOption(overwriteOutput) .withOption(helpOpt).withOption(sequentialAccessVectorOpt).withOption(namedVectorOpt) .withOption(logNormalizeOpt).withOption(labelMDOpt).create(); CommandLine cmdLine = null; try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } if (!cmdLine.hasOption(inputDirOpt)) { CommandLineUtil.printHelp(group); return -1; } if (!cmdLine.hasOption(outputDirOpt)) { CommandLineUtil.printHelp(group); return -1; } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = DefaultAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it // if // you can't instantiate it ClassUtils.instantiateAs(analyzerClass, Analyzer.class); } String type = null; String featureName = ""; if (cmdLine.hasOption(typeNameOpt)) { type = cmdLine.getValue(typeNameOpt).toString(); Object tempFN = cmdLine.getValue(featureNameOpt); if (tempFN != null) { featureName = tempFN.toString(); log.info("Getting tokens from " + type + "." + featureName.toString()); } else log.info("Getting tokens from " + type); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } String labelMDKey = null; if (cmdLine.hasOption(labelMDOpt)) { labelMDKey = cmdLine.getValue(labelMDOpt).toString(); } Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); // no annotation type degfin if (type != null) { BehemothDocumentProcessor.tokenizeDocuments(inputDir, type, featureName, tokenizedPath); } // no annotation type defined : rely on Lucene's analysers else { BehemothDocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); } boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; try { if (!processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; // if we are pruning by std dev, then // this will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); long vectorCount = docFrequenciesFeatures.getFirst()[1]; maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } // dump labels? if (labelMDKey != null) { conf.set(BehemothDocumentProcessor.MD_LABEL, labelMDKey); BehemothDocumentProcessor.dumpLabels(inputDir, new Path(outputDir, "labels"), conf); } } catch (RuntimeException e) { log.error("Exception caught", e); return -1; } return 0; }
From source file:com.elex.dmp.vectorizer.DictionaryVectorizer.java
License:Apache License
/** * Create Term Frequency (Tf) Vectors from the input set of documents in {@link SequenceFile} format. This * tries to fix the maximum memory used by the feature chunk per node thereby splitting the process across * multiple map/reduces.//from w ww .j av a 2 s . c o m * * @param input * input directory of the documents in {@link SequenceFile} format * @param output * output directory where {@link org.apache.mahout.math.RandomAccessSparseVector}'s of the document * are generated * @param tfVectorsFolderName * The name of the folder in which the final output vectors will be stored * @param baseConf * job configuration * @param normPower * L_p norm to be computed * @param logNormalize * whether to use log normalization * @param minSupport * the minimum frequency of the feature in the entire corpus to be considered for inclusion in the * sparse vector * @param maxNGramSize * 1 = unigram, 2 = unigram and bigram, 3 = unigram, bigram and trigram * @param minLLRValue * minValue of log likelihood ratio to used to prune ngrams * @param chunkSizeInMegabytes * the size in MB of the feature => id chunk to be kept in memory at each node during Map/Reduce * stage. Its recommended you calculated this based on the number of cores and the free memory * available to you per node. Say, you have 2 cores and around 1GB extra memory to spare we * recommend you use a split size of around 400-500MB so that two simultaneous reducers can create * partial vectors without thrashing the system due to increased swapping */ public static void createTermFrequencyVectors(Path input, Path output, String tfVectorsFolderName, Configuration baseConf, int minSupport, int maxNGramSize, float minLLRValue, float normPower, boolean logNormalize, int numReducers, int chunkSizeInMegabytes, boolean sequentialAccess, boolean namedVectors) throws IOException, InterruptedException, ClassNotFoundException { Preconditions.checkArgument(normPower == PartialVectorMerger.NO_NORMALIZING || normPower >= 0, "If specified normPower must be nonnegative", normPower); Preconditions.checkArgument( normPower == PartialVectorMerger.NO_NORMALIZING || (normPower > 1 && !Double.isInfinite(normPower)) || !logNormalize, "normPower must be > 1 and not infinite if log normalization is chosen", normPower); if (chunkSizeInMegabytes < MIN_CHUNKSIZE) { chunkSizeInMegabytes = MIN_CHUNKSIZE; } else if (chunkSizeInMegabytes > MAX_CHUNKSIZE) { // 10GB chunkSizeInMegabytes = MAX_CHUNKSIZE; } if (minSupport < 0) { minSupport = DEFAULT_MIN_SUPPORT; } Path dictionaryJobPath = new Path(output, DICTIONARY_JOB_FOLDER); int[] maxTermDimension = new int[1]; List<Path> dictionaryChunks; if (maxNGramSize == 1) { startWordCounting(input, dictionaryJobPath, baseConf, minSupport); dictionaryChunks = createDictionaryChunks(dictionaryJobPath, output, baseConf, chunkSizeInMegabytes, maxTermDimension); } else { CollocDriver.generateAllGrams(input, dictionaryJobPath, baseConf, maxNGramSize, minSupport, minLLRValue, numReducers); dictionaryChunks = createDictionaryChunks( new Path(new Path(output, DICTIONARY_JOB_FOLDER), CollocDriver.NGRAM_OUTPUT_DIRECTORY), output, baseConf, chunkSizeInMegabytes, maxTermDimension); } int partialVectorIndex = 0; Collection<Path> partialVectorPaths = Lists.newArrayList(); for (Path dictionaryChunk : dictionaryChunks) { Path partialVectorOutputPath = new Path(output, VECTOR_OUTPUT_FOLDER + partialVectorIndex++); partialVectorPaths.add(partialVectorOutputPath); makePartialVectors(input, baseConf, maxNGramSize, dictionaryChunk, partialVectorOutputPath, maxTermDimension[0], sequentialAccess, namedVectors, numReducers); } Configuration conf = new Configuration(baseConf); Path outputDir = new Path(output, tfVectorsFolderName); PartialVectorMerger.mergePartialVectors(partialVectorPaths, outputDir, conf, normPower, logNormalize, maxTermDimension[0], sequentialAccess, namedVectors, numReducers); HadoopUtil.delete(conf, partialVectorPaths); }
From source file:com.elex.dmp.vectorizer.FixDictionaryVectorizer.java
License:Apache License
/** * Create Term Frequency (Tf) Vectors from the input set of documents in {@link SequenceFile} format. This * tries to fix the maximum memory used by the feature chunk per node thereby splitting the process across * multiple map/reduces./* w w w . j a va 2s . com*/ * * @param input * input directory of the documents in {@link SequenceFile} format * @param output * output directory where {@link org.apache.mahout.math.RandomAccessSparseVector}'s of the document * are generated * @param tfVectorsFolderName * The name of the folder in which the final output vectors will be stored * @param baseConf * job configuration * @param normPower * L_p norm to be computed * @param logNormalize * whether to use log normalization * @param minSupport * the minimum frequency of the feature in the entire corpus to be considered for inclusion in the * sparse vector * @param maxNGramSize * 1 = unigram, 2 = unigram and bigram, 3 = unigram, bigram and trigram * @param minLLRValue * minValue of log likelihood ratio to used to prune ngrams * @param chunkSizeInMegabytes * the size in MB of the feature => id chunk to be kept in memory at each node during Map/Reduce * stage. Its recommended you calculated this based on the number of cores and the free memory * available to you per node. Say, you have 2 cores and around 1GB extra memory to spare we * recommend you use a split size of around 400-500MB so that two simultaneous reducers can create * partial vectors without thrashing the system due to increased swapping */ public static void createTermFrequencyVectors(Path input, Path output, String tfVectorsFolderName, Configuration baseConf, int minSupport, int maxNGramSize, float minLLRValue, float normPower, boolean logNormalize, int numReducers, int chunkSizeInMegabytes, boolean sequentialAccess, boolean namedVectors) throws IOException, InterruptedException, ClassNotFoundException { Preconditions.checkArgument(normPower == PartialVectorMerger.NO_NORMALIZING || normPower >= 0, "If specified normPower must be nonnegative", normPower); Preconditions.checkArgument( normPower == PartialVectorMerger.NO_NORMALIZING || (normPower > 1 && !Double.isInfinite(normPower)) || !logNormalize, "normPower must be > 1 and not infinite if log normalization is chosen", normPower); if (chunkSizeInMegabytes < MIN_CHUNKSIZE) { chunkSizeInMegabytes = MIN_CHUNKSIZE; } else if (chunkSizeInMegabytes > MAX_CHUNKSIZE) { // 10GB chunkSizeInMegabytes = MAX_CHUNKSIZE; } if (minSupport < 0) { minSupport = DEFAULT_MIN_SUPPORT; } //??? Path dictFilePath = new Path(PropertiesUtil.getBackUpDir(), DICTIONARY_FILE_FOLDER); Path dictionaryJobPath = new Path(output, DICTIONARY_JOB_FOLDER); int[] maxTermDimension = new int[1]; List<Path> dictionaryChunks; if (maxNGramSize == 1) { startWordCounting(input, dictionaryJobPath, baseConf, minSupport); dictionaryChunks = createDictionaryChunks(dictFilePath, output, baseConf, chunkSizeInMegabytes, maxTermDimension); } else { CollocDriver.generateAllGrams(input, dictionaryJobPath, baseConf, maxNGramSize, minSupport, minLLRValue, numReducers); //?????ngram? dictionaryChunks = createDictionaryChunks(dictFilePath, output, baseConf, chunkSizeInMegabytes, maxTermDimension); } int partialVectorIndex = 0; Collection<Path> partialVectorPaths = Lists.newArrayList(); for (Path dictionaryChunk : dictionaryChunks) { Path partialVectorOutputPath = new Path(output, VECTOR_OUTPUT_FOLDER + partialVectorIndex++); partialVectorPaths.add(partialVectorOutputPath); makePartialVectors(input, baseConf, maxNGramSize, dictionaryChunk, partialVectorOutputPath, maxTermDimension[0], sequentialAccess, namedVectors, numReducers); } Configuration conf = new Configuration(baseConf); Path outputDir = new Path(output, tfVectorsFolderName); PartialVectorMerger.mergePartialVectors(partialVectorPaths, outputDir, conf, normPower, logNormalize, maxTermDimension[0], sequentialAccess, namedVectors, numReducers); HadoopUtil.delete(conf, partialVectorPaths); }
From source file:com.elex.dmp.vectorizer.SparseVectorsFromSequenceFiles.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(//w ww. j a va 2s . c om "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) of the document frequencies of these vectors." + " Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less then 0 no vectors " + "will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = DefaultAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it ClassUtils.instantiateAs(analyzerClass, Analyzer.class); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); //TODO: move this into DictionaryVectorizer , and then fold SparseVectorsFrom with EncodedVectorsFrom to have one framework for all of this. DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; if (!processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); long vectorCount = docFrequenciesFeatures.getFirst()[1]; maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:com.elex.dmp.vectorizer.TFVectorsUseFixedDictionary.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(// www .j a v a 2 s . c o m "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) of the document frequencies of these vectors." + " Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less then 0 no vectors " + "will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = DefaultAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it ClassUtils.instantiateAs(analyzerClass, Analyzer.class); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); //TODO: move this into DictionaryVectorizer , and then fold SparseVectorsFrom with EncodedVectorsFrom to have one framework for all of this. DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? FixDictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : FixDictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; if (!processIdf) { FixDictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { FixDictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); long vectorCount = docFrequenciesFeatures.getFirst()[1]; maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, FixDictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, FixDictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf( new Path(outputDir, FixDictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:com.ml.hadoop.nlp.SparseVectorsFromSequenceFiles.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option dictionaryPathOpt = obuilder.withLongName("dictionaryPath") .withArgument(abuilder.withName("dictionaryPath").withMinimum(1).withMaximum(1).create()) .withDescription("Dictionary path for update TFIDF").withShortName("dp").create(); Option docFrequencyPathOpt = obuilder.withLongName("docFrequencyPath") .withArgument(abuilder.withName("docFrequencyPath").withMinimum(1).withMaximum(1).create()) .withDescription("Doc frequency path for update TFIDF").withShortName("dfp").create(); Option tfVectorsPathOpt = obuilder.withLongName("tfVectorsPath") .withArgument(abuilder.withName("tfVectorsPath").withMinimum(1).withMaximum(1).create()) .withDescription("TF Vectors path").withShortName("tfvp").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF , TFIDF or TFIDF_UPDATE") .withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(//from w ww. ja v a 2 s . c om "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, " + "it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) " + "of the document frequencies of these vectors. Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less " + "than 0 no vectors will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(dictionaryPathOpt).withOption(docFrequencyPathOpt).withOption(tfVectorsPathOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Changed... Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = StandardAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it AnalyzerUtils.createAnalyzer(analyzerClass); } //default process tfidf:1, tf:2, update tfidf:3 int processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = 2; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = 1; } else if ("tfidf_update".equalsIgnoreCase(wString)) { processIdf = 3; } else { throw new OptionException(weightOpt); } } else { processIdf = 1; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } log.info("Tokenizing documents in {}", inputDir); Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0 || maxDFPercent > 0.00; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; log.info("Creating Term Frequency Vectors, prune {}", shouldPrune); String dictionaryPath = null; if (cmdLine.hasOption(dictionaryPathOpt)) { dictionaryPath = (String) cmdLine.getValue(dictionaryPathOpt); log.info("begin dic path {}", dictionaryPath); } if (processIdf == 1) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else if (processIdf == 3) { log.info("begin update term----------------"); DictionaryVectorizer.createUpdateTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, maxNGramSize, dictionaryPath, norm, logNormalize, reduceTasks, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } String docFrequencyPaths = null; if (cmdLine.hasOption(dictionaryPathOpt)) { docFrequencyPaths = (String) cmdLine.getValue(docFrequencyPathOpt); log.info("doc frequency path {}", docFrequencyPaths); } String tfVectorsPaths = null; if (cmdLine.hasOption(tfVectorsPathOpt)) { tfVectorsPaths = (String) cmdLine.getValue(tfVectorsPathOpt); log.info("tf vectors path {}", tfVectorsPaths); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (processIdf == 1) { log.info("Calculating IDF"); docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); log.info("...docFrequencyPathBase {}, docFrequencyFile {}", docFrequenciesFeatures.getFirst()[0], docFrequenciesFeatures.getFirst()[1]); } else if (processIdf == 3) { // load docFrequency path List<Path> docFrequencyChunks = Lists.newArrayList(); String[] paths = docFrequencyPaths.split(","); long featureCount = 0; for (String path : paths) { int splitPos = path.lastIndexOf("/"); String docFrequencyPathBase = path.substring(0, splitPos); String docFrequencyFile = path.substring(splitPos + 1, path.length()); log.info("docFrequencyPathBase {}, docFrequencyFile {}", docFrequencyPathBase, docFrequencyFile); Path docFrequencyPath = new Path(docFrequencyPathBase, docFrequencyFile); docFrequencyChunks.add(docFrequencyPath); /*for (Pair<IntWritable, LongWritable> record : new SequenceFileIterable<IntWritable, LongWritable>(docFrequencyPath, true, conf)) { featureCount = Math.max(record.getFirst().get(), featureCount); }*/ } featureCount = 107623; featureCount++; long vectorCount = Long.MAX_VALUE; /*Path tfDirPath = new Path(tfVectorsPaths + "/part-r-00000"); int i = 0; for (Pair<Text, VectorWritable> record : new SequenceFileIterable<Text, VectorWritable>(tfDirPath, true, conf)) { i++; } if (i > 0) { vectorCount = i; }*/ vectorCount = 80000; //read docFrequencyFile to get featureCount and vectorCount Long[] counts = { featureCount, vectorCount }; log.info("featureCount {}, vectorCount------------------ {}", featureCount, vectorCount); docFrequenciesFeatures = new Pair<Long[], List<Path>>(counts, docFrequencyChunks); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { long vectorCount = docFrequenciesFeatures.getFirst()[1]; if (maxDFSigma >= 0.0) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); } long maxDFThreshold = (long) (vectorCount * (maxDF / 100.0f)); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); log.info("Pruning"); if (processIdf == 1 || processIdf == 3) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDFThreshold, minDf, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDFThreshold, minDf, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf == 1 || processIdf == 3) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:com.netease.news.vectorizer.SparseVectorsFromSequenceFiles.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(//from w w w.j av a2 s .c o m "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, " + "it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) " + "of the document frequencies of these vectors. Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less " + "than 0 no vectors will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = IKAnalyzer.class; // Class<? extends Analyzer> analyzerClass = StandardAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it AnalyzerUtils.createAnalyzer(analyzerClass); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } log.info("Tokenizing documents in {}", inputDir); Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); //TODO: move this into DictionaryVectorizer , and then fold SparseVectorsFrom with EncodedVectorsFrom // to have one framework for all of this. DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0 || maxDFPercent > 0.00; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; log.info("Creating Term Frequency Vectors"); if (processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { log.info("Calculating IDF"); docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { long vectorCount = docFrequenciesFeatures.getFirst()[1]; if (maxDFSigma >= 0.0) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); } long maxDFThreshold = (long) (vectorCount * (maxDF / 100.0f)); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); log.info("Pruning"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDFThreshold, minDf, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDFThreshold, minDf, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:edu.indiana.d2i.htrc.io.SparseVectorsFromTokenizedDoc.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(// w ww .jav a 2 s.c o m "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) of the document frequencies of these vectors." + " Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less then 0 no vectors " + "will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = DefaultAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it // if // you can't instantiate it ClassUtils.instantiateAs(analyzerClass, Analyzer.class); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } /* modification starts here */ Configuration conf = getConf(); // Path tokenizedPath = new Path(outputDir, // DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); // DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, // tokenizedPath, conf); Path tokenizedPath = inputDir; /* end modification */ boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; if (!processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent;// if we are pruning by std dev, then this // will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0D, conf); maxDF = (int) (maxDFSigma * stdDev); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }
From source file:mahout.vectorizer.SparseVectorsFromSequenceFiles.java
License:Apache License
@Override public int run(String[] args) throws Exception { DefaultOptionBuilder obuilder = new DefaultOptionBuilder(); ArgumentBuilder abuilder = new ArgumentBuilder(); GroupBuilder gbuilder = new GroupBuilder(); Option inputDirOpt = DefaultOptionCreator.inputOption().create(); Option outputDirOpt = DefaultOptionCreator.outputOption().create(); Option minSupportOpt = obuilder.withLongName("minSupport") .withArgument(abuilder.withName("minSupport").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Minimum Support. Default Value: 2").withShortName("s").create(); Option analyzerNameOpt = obuilder.withLongName("analyzerName") .withArgument(abuilder.withName("analyzerName").withMinimum(1).withMaximum(1).create()) .withDescription("The class name of the analyzer").withShortName("a").create(); Option chunkSizeOpt = obuilder.withLongName("chunkSize") .withArgument(abuilder.withName("chunkSize").withMinimum(1).withMaximum(1).create()) .withDescription("The chunkSize in MegaBytes. 100-10000 MB").withShortName("chunk").create(); Option weightOpt = obuilder.withLongName("weight").withRequired(false) .withArgument(abuilder.withName("weight").withMinimum(1).withMaximum(1).create()) .withDescription("The kind of weight to use. Currently TF or TFIDF").withShortName("wt").create(); Option minDFOpt = obuilder.withLongName("minDF").withRequired(false) .withArgument(abuilder.withName("minDF").withMinimum(1).withMaximum(1).create()) .withDescription("The minimum document frequency. Default is 1").withShortName("md").create(); Option maxDFPercentOpt = obuilder.withLongName("maxDFPercent").withRequired(false) .withArgument(abuilder.withName("maxDFPercent").withMinimum(1).withMaximum(1).create()) .withDescription(//from w w w.j a va 2 s .c o m "The max percentage of docs for the DF. Can be used to remove really high frequency terms." + " Expressed as an integer between 0 and 100. Default is 99. If maxDFSigma is also set, " + "it will override this value.") .withShortName("x").create(); Option maxDFSigmaOpt = obuilder.withLongName("maxDFSigma").withRequired(false) .withArgument(abuilder.withName("maxDFSigma").withMinimum(1).withMaximum(1).create()) .withDescription( "What portion of the tf (tf-idf) vectors to be used, expressed in times the standard deviation (sigma) " + "of the document frequencies of these vectors. Can be used to remove really high frequency terms." + " Expressed as a double value. Good value to be specified is 3.0. In case the value is less than 0 " + "no vectors will be filtered out. Default is -1.0. Overrides maxDFPercent") .withShortName("xs").create(); Option minLLROpt = obuilder.withLongName("minLLR").withRequired(false) .withArgument(abuilder.withName("minLLR").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional)The minimum Log Likelihood Ratio(Float) Default is " + LLRReducer.DEFAULT_MIN_LLR) .withShortName("ml").create(); Option numReduceTasksOpt = obuilder.withLongName("numReducers") .withArgument(abuilder.withName("numReducers").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) Number of reduce tasks. Default Value: 1").withShortName("nr") .create(); Option powerOpt = obuilder.withLongName("norm").withRequired(false) .withArgument(abuilder.withName("norm").withMinimum(1).withMaximum(1).create()) .withDescription( "The norm to use, expressed as either a float or \"INF\" if you want to use the Infinite norm. " + "Must be greater or equal to 0. The default is not to normalize") .withShortName("n").create(); Option logNormalizeOpt = obuilder.withLongName("logNormalize").withRequired(false) .withDescription("(Optional) Whether output vectors should be logNormalize. If set true else false") .withShortName("lnorm").create(); Option maxNGramSizeOpt = obuilder.withLongName("maxNGramSize").withRequired(false) .withArgument(abuilder.withName("ngramSize").withMinimum(1).withMaximum(1).create()) .withDescription("(Optional) The maximum size of ngrams to create" + " (2 = bigrams, 3 = trigrams, etc) Default Value:1") .withShortName("ng").create(); Option sequentialAccessVectorOpt = obuilder.withLongName("sequentialAccessVector").withRequired(false) .withDescription( "(Optional) Whether output vectors should be SequentialAccessVectors. If set true else false") .withShortName("seq").create(); Option namedVectorOpt = obuilder.withLongName("namedVector").withRequired(false) .withDescription("(Optional) Whether output vectors should be NamedVectors. If set true else false") .withShortName("nv").create(); Option overwriteOutput = obuilder.withLongName("overwrite").withRequired(false) .withDescription("If set, overwrite the output directory").withShortName("ow").create(); Option helpOpt = obuilder.withLongName("help").withDescription("Print out help").withShortName("h") .create(); Group group = gbuilder.withName("Options").withOption(minSupportOpt).withOption(analyzerNameOpt) .withOption(chunkSizeOpt).withOption(outputDirOpt).withOption(inputDirOpt).withOption(minDFOpt) .withOption(maxDFSigmaOpt).withOption(maxDFPercentOpt).withOption(weightOpt).withOption(powerOpt) .withOption(minLLROpt).withOption(numReduceTasksOpt).withOption(maxNGramSizeOpt) .withOption(overwriteOutput).withOption(helpOpt).withOption(sequentialAccessVectorOpt) .withOption(namedVectorOpt).withOption(logNormalizeOpt).create(); try { Parser parser = new Parser(); parser.setGroup(group); parser.setHelpOption(helpOpt); CommandLine cmdLine = parser.parse(args); if (cmdLine.hasOption(helpOpt)) { CommandLineUtil.printHelp(group); return -1; } Path inputDir = new Path((String) cmdLine.getValue(inputDirOpt)); Path outputDir = new Path((String) cmdLine.getValue(outputDirOpt)); int chunkSize = 100; if (cmdLine.hasOption(chunkSizeOpt)) { chunkSize = Integer.parseInt((String) cmdLine.getValue(chunkSizeOpt)); } int minSupport = 2; if (cmdLine.hasOption(minSupportOpt)) { String minSupportString = (String) cmdLine.getValue(minSupportOpt); minSupport = Integer.parseInt(minSupportString); } int maxNGramSize = 1; if (cmdLine.hasOption(maxNGramSizeOpt)) { try { maxNGramSize = Integer.parseInt(cmdLine.getValue(maxNGramSizeOpt).toString()); } catch (NumberFormatException ex) { log.warn("Could not parse ngram size option"); } } log.info("Maximum n-gram size is: {}", maxNGramSize); if (cmdLine.hasOption(overwriteOutput)) { HadoopUtil.delete(getConf(), outputDir); } float minLLRValue = LLRReducer.DEFAULT_MIN_LLR; if (cmdLine.hasOption(minLLROpt)) { minLLRValue = Float.parseFloat(cmdLine.getValue(minLLROpt).toString()); } log.info("Minimum LLR value: {}", minLLRValue); int reduceTasks = 1; if (cmdLine.hasOption(numReduceTasksOpt)) { reduceTasks = Integer.parseInt(cmdLine.getValue(numReduceTasksOpt).toString()); } log.info("Number of reduce tasks: {}", reduceTasks); Class<? extends Analyzer> analyzerClass = StandardAnalyzer.class; if (cmdLine.hasOption(analyzerNameOpt)) { String className = cmdLine.getValue(analyzerNameOpt).toString(); analyzerClass = Class.forName(className).asSubclass(Analyzer.class); // try instantiating it, b/c there isn't any point in setting it if // you can't instantiate it AnalyzerUtils.createAnalyzer(analyzerClass); } boolean processIdf; if (cmdLine.hasOption(weightOpt)) { String wString = cmdLine.getValue(weightOpt).toString(); if ("tf".equalsIgnoreCase(wString)) { processIdf = false; } else if ("tfidf".equalsIgnoreCase(wString)) { processIdf = true; } else { throw new OptionException(weightOpt); } } else { processIdf = true; } int minDf = 1; if (cmdLine.hasOption(minDFOpt)) { minDf = Integer.parseInt(cmdLine.getValue(minDFOpt).toString()); } int maxDFPercent = 99; if (cmdLine.hasOption(maxDFPercentOpt)) { maxDFPercent = Integer.parseInt(cmdLine.getValue(maxDFPercentOpt).toString()); } double maxDFSigma = -1.0; if (cmdLine.hasOption(maxDFSigmaOpt)) { maxDFSigma = Double.parseDouble(cmdLine.getValue(maxDFSigmaOpt).toString()); } float norm = PartialVectorMerger.NO_NORMALIZING; if (cmdLine.hasOption(powerOpt)) { String power = cmdLine.getValue(powerOpt).toString(); if ("INF".equals(power)) { norm = Float.POSITIVE_INFINITY; } else { norm = Float.parseFloat(power); } } boolean logNormalize = false; if (cmdLine.hasOption(logNormalizeOpt)) { logNormalize = true; } Configuration conf = getConf(); Path tokenizedPath = new Path(outputDir, DocumentProcessor.TOKENIZED_DOCUMENT_OUTPUT_FOLDER); //TODO: move this into DictionaryVectorizer , and then fold SparseVectorsFrom with EncodedVectorsFrom // to have one framework for all of this. DocumentProcessor.tokenizeDocuments(inputDir, analyzerClass, tokenizedPath, conf); boolean sequentialAccessOutput = false; if (cmdLine.hasOption(sequentialAccessVectorOpt)) { sequentialAccessOutput = true; } boolean namedVectors = false; if (cmdLine.hasOption(namedVectorOpt)) { namedVectors = true; } boolean shouldPrune = maxDFSigma >= 0.0; String tfDirName = shouldPrune ? DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-toprune" : DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER; if (processIdf) { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, -1.0f, false, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } else { DictionaryVectorizer.createTermFrequencyVectors(tokenizedPath, outputDir, tfDirName, conf, minSupport, maxNGramSize, minLLRValue, norm, logNormalize, reduceTasks, chunkSize, sequentialAccessOutput, namedVectors); } Pair<Long[], List<Path>> docFrequenciesFeatures = null; // Should document frequency features be processed if (shouldPrune || processIdf) { docFrequenciesFeatures = TFIDFConverter.calculateDF(new Path(outputDir, tfDirName), outputDir, conf, chunkSize); } long maxDF = maxDFPercent; //if we are pruning by std dev, then this will get changed if (shouldPrune) { Path dfDir = new Path(outputDir, TFIDFConverter.WORDCOUNT_OUTPUT_FOLDER); Path stdCalcDir = new Path(outputDir, HighDFWordsPruner.STD_CALC_DIR); // Calculate the standard deviation double stdDev = BasicStats.stdDevForGivenMean(dfDir, stdCalcDir, 0.0, conf); long vectorCount = docFrequenciesFeatures.getFirst()[1]; maxDF = (int) (100.0 * maxDFSigma * stdDev / vectorCount); // Prune the term frequency vectors Path tfDir = new Path(outputDir, tfDirName); Path prunedTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER); Path prunedPartialTFDir = new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER + "-partial"); if (processIdf) { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, -1.0f, false, reduceTasks); } else { HighDFWordsPruner.pruneVectors(tfDir, prunedTFDir, prunedPartialTFDir, maxDF, conf, docFrequenciesFeatures, norm, logNormalize, reduceTasks); } HadoopUtil.delete(new Configuration(conf), tfDir); } if (processIdf) { TFIDFConverter.processTfIdf(new Path(outputDir, DictionaryVectorizer.DOCUMENT_VECTOR_OUTPUT_FOLDER), outputDir, conf, docFrequenciesFeatures, minDf, maxDF, norm, logNormalize, sequentialAccessOutput, namedVectors, reduceTasks); } } catch (OptionException e) { log.error("Exception", e); CommandLineUtil.printHelp(group); } return 0; }